Skip to content
Unet model for classifying objects on Satellite Images
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.


This project focuses on using a modified Unet model for image classification of satellite images. The original data for this project is from the DSTL Kaggle Challenge The data for this project has been modified to only use the 8 band Multispectral Tiff collection and the classes have been reduced to 5:

  • Buildings
  • Roads and Tracks
  • Trees
  • Crops
  • Water

The model was trained on 24 images, each with 8 bands and 5 classification masks. Image patches were augmented using ndimage and matrix math.

Training and prediction was completed on a NVIDIA P100 GPU with 120GB of RAM

Libraries: Keras, Tensorflow, Tifffile, Python3, numpy, scipy

You can’t perform that action at this time.